How to Run an eDiscovery Vendor Selection in Ten Steps, From Data Map to Signed MSA
Most eDiscovery purchases go wrong in three places: data that can't legally leave its country, quotes that can't be compared, and contracts that make leaving expensive. These ten steps deal with all three.
Steps 1–2: Profile Your Matters and Map Where the Data Lives
Most eDiscovery vendor selection starts with a demo. It should start with a spreadsheet. If you can't describe what you're buying for, you can't judge fit, and you'll pay for features sized to someone else's caseload.
Step 1: Build a matter profile
Go back over two or three years of litigation, investigations, regulatory requests and internal reviews. For each matter, record:
- volume collected
- volume still hosted after culling
- how long it stayed hosted
- peak reviewer headcount
- languages involved
Most teams find their work falls into a few patterns. A common one is a steady run of matters under 50 GB plus one or two a year above 500 GB. That pattern matters more than any feature list because it determines which pricing model is cheapest for you.
Step 2: Map the data sources and jurisdictions
For each source, list the system (Microsoft 365, Google Workspace, Slack, Teams chat, mobile devices, ERP exports, shared drives), the custodians who use it, and the country where the data is stored and where those custodians are based. Flag anything awkward: regional chat apps, short-message formats, scanned documents in non-Latin scripts and engineering file types. You'll reuse this map for the RFP requirements, the proof-of-concept sample and the transfer analysis, so finish it before anyone talks to a sales team.
Step 3: Settle Cross-Border Transfer Limits Before You Shortlist
Data residency rules out more vendors than any missing feature does, so settle it first. Transfer rules vary widely by jurisdiction. Some limit the act of handing evidence to a foreign court or regulator, not just where the data is stored.
- EU and UK GDPR: Sending personal data outside the EEA or UK needs a transfer mechanism. That's usually an adequacy decision, or standard contractual clauses backed by a transfer impact assessment. There is a derogation for establishing or defending legal claims, but regulators read it narrowly. Plan to cull and pseudonymize in-region before anything leaves. Some countries, France among them, also have blocking statutes that can affect foreign productions.
- China's PIPL: Exporting personal information generally requires one of three routes: a regulator security assessment, a filed standard contract or certification. Which one applies depends on volume and sensitivity. A separate requirement calls for approval from the competent authorities before personal information stored in China goes to foreign judicial or enforcement bodies. In practice, processing and reviewing in-country is often the only workable route.
- India's DPDP Act: The Act generally permits transfers except to countries the government restricts. Its implementing rules are being phased in. Sectoral rules, such as localization requirements in financial services, can be stricter and need checking on their own.
- Elsewhere: Saudi Arabia's PDPL, the UAE's federal data protection law (plus separate regimes in some free zones) and Brazil's LGPD each set their own transfer conditions.
In practice, every jurisdiction on your data map needs either a vendor-operated hosting region in or near it, or a deployable on-premises or private-cloud option. RelativityOne, Everlaw, DISCO, Reveal, Nuix and the large service providers all have different regional coverage. Ask each for its actual list of data center locations, because a map on a marketing page isn't enough. Before collection, have privacy counsel in each affected jurisdiction confirm the approach.
Steps 4–6: Shortlist, Issue the RFP and Put Every Quote on One Basis
Step 4: Shortlist three to five vendors that pass the residency filter and handle your source types. Step 5: Send a short RFP built from your matter profile. Ask for pricing against two or three defined scenarios rather than a rate card. Also ask exactly how billable volume is measured:
- compressed or expanded
- before or after deduplication
- monthly peak or monthly average
Step 6 is where most comparisons fall apart. Vendors quote per GB, per user or as a flat subscription, and you can't compare those headline rates directly. Instead, price every vendor against the same matter and compare totals.
| Model | Typical 2026 range (indicative) | Cheapest when |
|---|---|---|
| Per-GB hosting | Roughly $3–$20 per GB per month; processing from included to around $50 per GB | Small, short matters with many reviewers |
| Per-user | Roughly $40–$150 per user per month, plus lower hosting | Large data, small review teams |
| Subscription | Roughly $30,000 to $250,000+ per year for a volume tier | A steady flow of matters across the year |
These are typical 2026 market ranges. They vary by contract, region and currency, so confirm them with each vendor.
Hypothetical worked example: A 60-lawyer firm expects one matter with 500 GB collected, an average of 200 GB hosted for 12 months, 15 reviewers for six months and 3 admins all year.
- Vendor A charges $15/GB/month with processing and users included: $36,000.
- Vendor B charges $8/GB/month, $20/GB processing and $90/user/month: $19,200 + $10,000 + $8,100 + $3,240 = $40,540.
- Vendor C charges a flat $48,000 subscription for up to 1 TB hosted, with unlimited users.
B has the lowest hosting rate, yet it costs more than A. For one matter, A is cheapest. With two matters like this a year, and combined hosting under 1 TB, A rises to about $72,000 and B to about $81,000, while C stays at $48,000. Your matter profile decides the winner, not the rate card.
Add the line items that usually get left out: analytics and technology-assisted review (TAR) fees, translation, production or export charges, project management (often somewhere around $150–$275 per hour), cold-storage rates and end-of-matter data return.
Steps 7–8: Run a Scored Proof of Concept on Your Own Data
Vendor demo datasets are built to look good. Step 7 is a proof of concept on 20–50 GB of your own data that is either non-privileged or cleared for this use. Give every finalist the same sample and the same tasks over two to three weeks. Include the hard material from your data map, such as chat threads, mobile exports, foreign-language documents and unusual file types. Seed the sample with 30–50 known relevant documents so you can see whether search and analytics actually surface them.
| Criterion | Weight | What to measure |
|---|---|---|
| Processing fidelity | 25% | Exception rate, chat threading, metadata preserved |
| Search and analytics | 25% | Seeded documents found, TAR setup time |
| Reviewer experience | 20% | Documents per hour in a timed batch, coding errors |
| Production | 15% | Load file accuracy, redaction and privilege log output |
| Support | 15% | Response time and quality on two staged tickets |
Adjust the weights to suit your practice, but fix them before testing starts so nobody can tune them afterward to favor a vendor. Have at least two reviewers score each vendor independently.
Step 8 is security and commercial diligence, run in parallel with the POC. Ask each vendor for:
- a current SOC 2 Type II report or ISO 27001 certificate
- a recent penetration test summary
- SSO and MFA support
- encryption details, including any customer-managed key option
- the subprocessor list
- two references whose matter profile looks like yours
If a vendor won't commit in writing that your data won't be used to train its generative features, treat that as disqualifying.
Steps 9–10: Negotiate the MSA Terms That Decide Your Exit Cost
Step 9: Negotiate. Vendors usually move less on the rate card than on the terms around it. Concentrate on these:
- Volume definition: Bill on post-deduplication, expanded GB, measured as a monthly average rather than a peak.
- Price protection: Lock rates for the initial term and cap renewal increases, for example at a low single-digit percentage or a named inflation index.
- Data location: Name the hosting regions in the contract. Require advance notice and a right to object before any change of subprocessor or region.
- Exit: Require return of data in native format, with load files and full coding history, within a fixed window such as 30 days. There should be no export or egress fee, and the vendor should provide a certificate of deletion afterward.
- Security and liability: Require breach notice within a short, defined window (24–72 hours is a common ask) and a separate, higher liability cap for data breaches. Reject a general cap equal to one month of fees.
- Commitments: Make sure any minimum spend applies across all matters and unused volume rolls over.
Have a lawyer qualified in the governing-law jurisdiction review both the MSA and the data processing agreement. A clause that's standard in one country can be unenforceable in another.
Step 10: Sign with a written onboarding and exit plan attached. It should say who administers the platform, which pending matters move first, and, on one page, how you'd get everything back if you left. For the final decision, use this rule: if the top two POC scores are within five points, choose the vendor with the cheaper, faster exit.
Frequently asked questions
How long does an eDiscovery vendor selection usually take?
For a mid-sized firm or legal department, 8–14 weeks is realistic. That's roughly two to three weeks for data mapping and the RFP, two to three for the proof of concept, and three to six for negotiation, with diligence running alongside. Cross-border matters involving China or several EU countries often take longer because privacy counsel have to sign off on the transfer approach.
Should we choose a software platform or a managed service provider?
It mostly comes down to in-house litigation support staff. If you have one or two experienced eDiscovery specialists, a self-service platform usually gives better value. Without them, or with occasional very large matters, a provider that runs the platform for you is often the better choice. Many providers resell the same major platforms, so compare service quality and total pricing, not just the software.
Can we run a proof of concept on live client data?
Only with the client's consent, and only after a confidentiality agreement and data processing terms are signed with each vendor. Many teams use data from closed matters or internal corporate data that privacy counsel have cleared. Whatever you use, require every vendor to delete the sample after the POC and confirm the deletion in writing.
Is per-GB pricing always worse than a subscription?
No. Per-GB pricing is often cheaper for teams with a few small, short matters a year, because you aren't paying for unused capacity. Subscriptions tend to come out ahead once hosted volume stays fairly steady across several concurrent matters. Run your own matter profile through each structure before deciding.
Written by the Lexstack Cloud editorial team. Prices, features and rules change often - confirm details with the vendor or a licensed professional before you decide.